Repository navigation
docs: propose guest foreground work and restart notices - #74
vincentkoc wants to merge 4 commits into
Conversation
|
🦞👀 Pull request received. I will update this pull request when review starts. ClawSweeper review completeClawSweeper finished reviewing this revision. The review result is being finalized. |
|
Codex review: needs changes before merge. What this changesThis PR proposes bounded guest turns, persistent restart-interruption notices, fresh authorization to continue, and owner-controlled allocation cleanup. Example: A Gateway restarts while guest request guest-run-42 is executing.
Review scores
ProductKind: Docs · Worth it: Yes Merge readiness⛔ Needs changes before merge - 1 item remains Keep this PR open as a substantive, maintainer-authored draft RFC that is absent from main; no concrete patch defect was found. Priority: P3 Before merge
FindingsNone. Agent review detailsHow this fits togetherThis design repository records proposed contracts for OpenClaw; the RFC connects authenticated guest admission to execution, persisted recovery state, transcript notices, and sandbox retirement. flowchart TD
A[Authenticated guest request] --> B[Admission and current policy]
B --> C[Bounded foreground turn]
C --> D[Persisted run decision]
D --> E[Restart lifecycle settlement]
E --> F[Terminal outcome and thread notice]
C --> G[Original allocation owner]
G --> H[Verified shutdown and cleanup]
Technical reviewBest possible solution: Complete the repository's discussion and acceptance lifecycle, then implement the qualified guest policy through existing admission, lifecycle, storage, and sandbox owners. Do we have a high-confidence way to reproduce the issue? This PR proposes a new guest policy rather than repairing a reproduced repository bug; no target runtime was executed. Is this the best way to solve the issue? The design uses existing owners and explicitly addresses replay, current authorization, uncertain effects, containment, and rollback without claiming those implementation requirements are already proven. AGENTS.md: not found in the target repository. Codex review notes: model internal, reasoning medium; reviewed against 967d9aac7472. SecurityNone. EvidenceWhat I checked:
Likely related people:
LabelsLabel changes:
Label justifications:
Rating scale6/6 🦀 challenger crab · 5/6 🦞 diamond lobster · 4/6 🐚 platinum hermit · 3/6 🦐 gold shrimp · 2/6 🦪 silver shellfish · 1/6 🧂 unranked krab. Overall follows the weaker of proof and patch quality; ✨ marks media proof (a screenshot, video, or linked artifact) that directly shows the changed behavior. WorkflowClawSweeper edits this one comment on every review. Comment HistoryReview history (4 earlier review cycles)
Reviewed October 12, 2026, 10:40 AM ET / 14:40 UTC (Revision 5). |
|
@clawsweeper re-review Please review exact head 0e44342. The RFC remains draft pending maintainer discussion and acceptance. |
|
🦞🧹 I asked ClawSweeper to review this item again. Re-review progress:
|
What Problem This Solves
Guest users need a clear outcome when a Gateway restart interrupts their work. Automatically resuming that work requires authority and recovery machinery that the first guest experience does not need.
Why This Change Was Made
Propose bounded foreground guest turns, a persistent stopped-after-restart notice, and a fresh authorized request to continue in the same thread. Reuse existing session, transcript, and sandbox registry SQLite payloads: no new table, column, database, sidecar, or schema-version bump. The allocation marker
retirementPolicy: 'foreground-owner'keeps cleanup with its original owner until whole-process shutdown is confirmed.User Impact
This is a design proposal only. History remains available, interrupted guest work never resumes automatically, and staff behavior stays unchanged. Durable jobs and unqualified detached execution remain outside the initial guest profile. Generic cleanup retains marked or uncertain allocations. Older builds can ignore both markers, so incompatible downgrades require quiescing and reconciliation under compatible code first. Schema compatibility alone does not make rollback safe. Runtime containment and engine qualification remain implementation requirements.
Related proposals: #70 and #71.
Evidence
The required maintainer-discussion thread is pending. This RFC remains draft and unmerged, with no implementation issue until acceptance.